Pith. sign in

REVIEW 3 major objections 6 minor 93 references

PeftCD: Leveraging Vision Foundation Models with Parameter-Efficient Fine-Tuning for Remote Sensing Change Detection

T0 review · 3 major / 6 minor · reviewed 2026-08-04 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper claims that a frozen vision foundation model adapted only with LoRA or Adapter modules plus a lightweight decoder reaches state-of-the-art accuracy on seven remote sensing change detection benchmarks while training as few as 2.86

desk verdict PeftCD's cross-dataset SOTA claim is built from four different model variants, so the headline overstates; the frozen-backbone + PEFT result itself is credible and worth a serious look. read the letter →

arxiv 2509.09572 v1 pith:ILXMGPVS submitted 2025-09-11 cs.CV

Sijun Dong , Yuxuan Hu , LiBo Wang , Geng Chen , Xiaoliang Meng This is my paper
classification cs.CV
keywords changedetectionremotesensingvisionfoundationmodelsparameter-efficientfine-tuningLoRAAdapterSAM2DINOv3
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

PeftCD is a change detection framework that freezes a large vision foundation model and trains only a small set of injected parameters. The paper's central claim is that this minimal adaptation—LoRA or Adapter modules on a frozen SAM2 or DINOv3 encoder, plus a lightweight decoder—is enough to match or beat the published state of the art on seven remote sensing change detection benchmarks. The authors report IoU scores of 73.81% on SYSU-CD, 92.05% on WHUCD, 64.07% on MSRSCD, 76.89% on MLCD, 97.01% on CDD, 52.25% on S2Looking, and 85.62% on LEVIR-CD, while training as few as 2.86 million parameters. If correct, this shows that strong pretrained vision representations can be adapted to change detection very cheaply, avoiding full fine-tuning of large backbones.

What carries the argument

The weight-sharing Siamese encoder is the core: one frozen VFM processes both temporal images, and every Transformer block carries either a LoRA low-rank update (A and B matrices on qkv projections, r=8, alpha=32) or a pre-block residual Adapter MLP (down/up projections, bottleneck 32). Only these PEFT weights are updated during training. Between the temporal streams, features from alternating layers are exchanged to create mixed-temporal representations. For the DINOv3 backbone—which outputs only single-scale 1/16 features—the MFCE decoder performs position-adaptive attention-weighted fusion of same-scale multi-layer features, ASPP-based context enhancement, and progressive upsampling to re

What would settle it

Re-run a top published competitor (for example CD-STMamba or ChangeMamba) under PeftCD's exact training schedule, crop size, batch size, and validation checkpoint selection on SYSU-CD and LEVIR-CD; if it reaches or exceeds the reported IoU, the state-of-the-art claim fails. Independently, train one fixed variant (say DINOv3+LoRA) on all seven datasets and check whether it still beats every published number, since the abstract's scores mix different per-dataset winners.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that a frozen vision foundation model equipped with parameter-efficient modules can carry remote sensing change detection on its own. With all backbone weights frozen, injecting LoRA into the self-attention qkv projections (rank 8, scale 32) or a pre-block residual Adapter MLP (bottleneck 32) into every Transformer block, the model outperforms dedicated change detection networks that train tens to hundreds of millions of parameters. The DINOv3 instantiation adds a decoder that fuses features from layers 5, 11, 17, and 23 with per-location attention weights, expands context with an ASPP module, and progressively upsamples; the SAM2 instantiation uses

Load-bearing premise

The central claim assumes it is fair to compare PeftCD's numbers against those copied from other papers despite differing training protocols, and to claim 'state of the art' while choosing a different PeftCD variant for each dataset.

Editorial extensions

If this is right

  • Replacing task-specific encoder-decoder networks with a frozen VFM plus PEFT is enough to set state-of-the-art numbers on multiple public change detection benchmarks.
  • Training only 2.86 million parameters (DINOv3+LoRA) makes high-accuracy change detection feasible in GPU- or memory-limited settings, since the frozen backbone requires no gradient updates.
  • PEFT consistently beats a fully frozen encoder on both SAM2 and DINOv3, so the injected modules are doing essential domain adaptation rather than contributing negligible tweaks.
  • The MFCE decoder's gains over a plain upsampling decoder indicate that ViT single-scale features need explicit multi-layer fusion and context expansion to support pixel-level boundary-accurate prediction.
  • The paradigm points toward adapting large-scale VFMs to other remote sensing tasks without full fine-tuning.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper's per-dataset SOTA picks the best variant (for example, SAM2+LoRA for LEVIR-CD 85.62 vs DINOv3+LoRA 85.32, and SAM2+Adapter for CDD); a single fixed variant across all seven datasets would be a stricter, untested claim.
  • The numbers quoted from prior work are taken from their original papers; because training protocols, image tiling, and checkpoint selection differ, a direct re-benchmark under PeftCD's exact settings would be needed to confirm the SOTA margins.
  • Boundary quality and pseudo-change suppression are argued mainly through qualitative figures and global IoU/F1; targeted metrics such as boundary IoU or per-region false-positive counts would quantify these specific advertised strengths.
  • It would be a natural next test to swap in other frozen foundation models (e.g., a CLIP ViT or a DINOv2) with the same PEFT recipe; the paper's interpretation predicts similar gains, but it does not run that experiment.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper proposes PeftCD, a change-detection framework that freezes a Vision Foundation Model (SAM2 or DINOv3) as a weight-sharing Siamese encoder and injects only LoRA or Adapter modules as trainable parameters. A lightweight decoder and a layer-exchange interaction are added for temporal modeling. The method is evaluated on seven remote sensing change-detection datasets. The central empirical claim is that PeftCD achieves state-of-the-art IoU on all seven benchmarks while training only 2.86M parameters in its most efficient variant.

Significance. If the claims were supported as stated, the paper would provide a practical and valuable result: frozen VFMs plus small PEFT modules can match or beat task-specific change-detection architectures at a fraction of the trainable parameters. The ablation in Table IX does give credible evidence that PEFT improves over frozen encoders on both backbones and most datasets, and the MFCE decoder ablation in Table X supports the decoder design. The release of code and pretrained models would aid reproducibility. However, the headline SOTA claim is weakened by the use of different PeftCD variants for different datasets and by reliance on externally reported baseline numbers without common-protocol reruns or uncertainty estimates.

major comments (3)
  1. [Abstract; §IV-E, Table IX; §V-C, Table XI] The headline claim that 'PeftCD achieves state-of-the-art performance across seven public datasets' aggregates results from four different configurations. Table IX shows that SYSU-CD, WHUCD, S2Looking, and MSRSCD are best with DINOv3+LoRA; MLCD is best with DINOv3+Adapter; CDD is best with SAM2+Adapter; and LEVIR-CD is best with SAM2+LoRA. The abstract's 2.86M trainable-parameter statement applies only to DINOv3+LoRA; Table XI reports 10.15M and 11.00M trainable parameters for the SAM2 variants. Thus no single fixed model attains all seven reported numbers, and the claim in the conclusion of an 'optimal balance of accuracy, efficiency, and generalization' rests on per-dataset oracle selection. The paper should report one fixed configuration's complete results as the primary claim and present per-configuration results separately.
  2. [§IV-E; Tables II-VIII] The comparison tables list baseline numbers copied from prior papers without re-running those baselines under the same protocol used for PeftCD. The reported gains over the strongest baselines are in several cases small (e.g., SYSU-CD IoU 73.81 vs. 73.45; S2Looking 52.25 vs. 51.51; LEVIR-CD 85.62 vs. 85.58), and no variance or significance measure is reported anywhere. Under different training schedules, crops, and hardware, such margins may not be reproducible. The paper should either re-run key baselines under the identical settings or provide error bars and statistical comparison before claiming to 'outperform all competing approaches.'
  3. [§III-B2; §IV-F] Layer exchange is introduced as a core component of PeftCD ('Feature Interaction via Layer Exchange', §III-B2), but its contribution is never ablated. The ablations in Tables IX and X vary the backbone, PEFT method, and decoder, but no experiment removes or alters the layer-exchange mechanism. Since layer exchange is adopted from EfficientCD [40], its effect within the new PEFT-based framework should be quantified to support the claim that it contributes to the change-representation improvement.
minor comments (6)
  1. [§III-B1] Typo: 'intergate' should be 'integrate'.
  2. [§IV-E] In the text, 'Convformer (51.51)' should be 'ConvFormer-CD (51.51)' for consistency with Table IV and the reference list.
  3. [Tables II-IX] Column ordering of datasets is inconsistent across tables (e.g., Table IX lists CDD last for SAM2 but earlier for DINOv3; Tables II-VIII use varying orders). A consistent dataset order would make cross-table comparison easier.
  4. [§IV-B] Implementation details do not state the number of training epochs for each dataset. This is needed to assess convergence and comparability with baselines.
  5. [§IV-B; §IV-A] For S2Looking, LEVIR-CD, and MSRSCD, the paper states random 256×256 cropping during training, but the listed image sizes are 1024×1024 or larger. The effective training resolution and its effect on performance should be stated more explicitly.
  6. [§V-A] The discussion candidly lists limitations of the single-scale ViT decoder (resolution bound, fixed ASPP dilation rates, memory cost). These limitations should be connected more explicitly to the SOTA claims, since they qualify the generalization statement in the conclusion.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation; the empirical SOTA claim is externally benchmarked, though the abstract aggregates results across four PeftCD variants (a reporting concern, not circularity).

full rationale

PeftCD's derivation chain is an empirical architecture study: frozen SAM2/DINOv3 backbones plus standard LoRA/Adapter modules (Eqs. 1-4), a layer-exchange interaction borrowed from the authors' prior EfficientCD [40], and lightweight decoders. Nothing in the method is defined in terms of the reported test metrics. The PEFT gains are shown by ablations against Frozen Encoder baselines (Table IX), and the decoder gains by ablations against a simplified decoder (Table X); neither ablation is fitted to the headline IoU values. The SOTA comparisons in Tables II-VIII use public test splits and prior published numbers, so the central claim is externally falsifiable. The only self-citation of note is EfficientCD [40] for layer exchange, but that is an applied design component, not a load-bearing theorem, and no uniqueness claim rests on it. The abstract's per-dataset best numbers come from different PeftCD variants (e.g., LEVIR-CD from SAM2+LoRA, SYSU-CD from DINOv3+LoRA), which is a clarity/comparability issue rather than a circular reduction; no equation or fitted parameter is being renamed as a prediction. Hence no significant circularity.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The paper's central claim is empirical, so the ledger mostly records hand-chosen PEFT hyperparameters and domain assumptions about transfer and comparability. No new physical or conceptual entities are introduced; the MFCE decoder is an architectural module, not an invented entity in the ledger sense.

free parameters (4)
  • LoRA rank r = 8
    Fixed for all experiments; standard default but no sensitivity analysis, so performance could depend on this choice.
  • LoRA scale alpha = 32
    Fixed without ablation; controls update magnitude and is a hand-chosen hyperparameter.
  • Adapter bottleneck dimension = 32
    Fixed for all Adapter variants; no ablation over bottleneck sizes.
  • DINOv3 layers used for fusion = 5, 11, 17, 23
    Hand-picked as analogous to ResNet stages; no ablation over alternative layer selections, and this choice directly feeds the MFCE decoder.
assumptions (3)
  • domain assumption A frozen VFM pretrained on natural images, adapted by tiny LoRA/Adapter modules, is sufficient for remote sensing change detection.
    The entire framework assumes small trainable modules can steer frozen backbones; Table IX supports this empirically but it is an a priori premise of the design.
  • domain assumption Numbers reported for baselines from prior papers are directly comparable to the authors' runs.
    SOTA claims in Section IV-E rely on this; the paper does not provide a shared harness, error bars, or a full re-run protocol for all baselines.
  • ad hoc to paper Layer exchange between temporal streams improves change representation.
    Adopted from the authors' prior EfficientCD work without an isolated ablation of this component inside PeftCD.

how reviews work

0 comments
Cite this review

Pith. "Pith review of PeftCD: Leveraging Vision Foundation Models with Parameter-Efficient Fine-Tuning for Remote Sensing Change Detection." pith.science (2026). https://pith.science/paper/ILXMGPVS

@misc{pith2026250909572,
  author       = {Pith},
  title        = {Pith review of: PeftCD: Leveraging Vision Foundation Models with Parameter-Efficient Fine-Tuning for Remote Sensing Change Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ILXMGPVS}},
  note         = {Machine review of arXiv:2509.09572}
}
read the original abstract

To tackle the prevalence of pseudo changes, the scarcity of labeled samples, and the difficulty of cross-domain generalization in multi-temporal and multi-source remote sensing imagery, we propose PeftCD, a change detection framework built upon Vision Foundation Models (VFMs) with Parameter-Efficient Fine-Tuning (PEFT). At its core, PeftCD employs a weight-sharing Siamese encoder derived from a VFM, into which LoRA and Adapter modules are seamlessly integrated. This design enables highly efficient task adaptation by training only a minimal set of additional parameters. To fully unlock the potential of VFMs, we investigate two leading backbones: the Segment Anything Model v2 (SAM2), renowned for its strong segmentation priors, and DINOv3, a state-of-the-art self-supervised representation learner. The framework is complemented by a deliberately lightweight decoder, ensuring the focus remains on the powerful feature representations from the backbones. Extensive experiments demonstrate that PeftCD achieves state-of-the-art performance across multiple public datasets, including SYSU-CD (IoU 73.81%), WHUCD (92.05%), MSRSCD (64.07%), MLCD (76.89%), CDD (97.01%), S2Looking (52.25%) and LEVIR-CD (85.62%), with notably precise boundary delineation and strong suppression of pseudo-changes. In summary, PeftCD presents an optimal balance of accuracy, efficiency, and generalization. It offers a powerful and scalable paradigm for adapting large-scale VFMs to real-world remote sensing change detection applications. The code and pretrained models will be released at https://github.com/dyzy41/PeftCD.

Figures

Figures reproduced from arXiv: 2509.09572 by the authors.

Figure 1
Figure 1. Architecture of PeftCD. Bi-temporal images are encoded by a shared VFM backbone with injected PEFT modules. Features undergo exchange before [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. SAM2CD: PeftCD with SAM2 Backbone. Features are exchanged between temporal streams and fused through FPN before dual-branch decoding [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Architecture of the PeftCD model based on DINOv3 (denoted as DINO3CD). [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Qualitative comparison between PeftCD and competing methods on the SYSU-CD dataset. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Qualitative comparison between PeftCD and competing methods on the WHUCD dataset. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Qualitative comparison between PeftCD and competing methods on the S2Looking dataset. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Qualitative comparison between PeftCD and competing methods on the MSRSCD dataset. [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: Qualitative comparison between PeftCD and competing methods on the MLCD dataset. [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: Qualitative comparison between PeftCD and competing methods on the CDD dataset. [PITH_FULL_IMAGE:figures/full_fig_p010_9.png]
Figure 10
Figure 10. Figure 10: Qualitative comparison between PeftCD and competing methods on the LEVIR-CD dataset. [PITH_FULL_IMAGE:figures/full_fig_p010_10.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

93 extracted references · 7 linked inside Pith

  1. [40]

    A CNN-transformer network with multiscale context aggregation for fine-grained cropland change detection,

    M. Liu, Z. Chai, H. Deng, and R. Liu, “A CNN-transformer network with multiscale context aggregation for fine-grained cropland change detection,” vol. 15, pp. 4297–4306

  2. [1]

    Segment anything,

    A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A. C. Berg, W.-Y . Lo, P. Doll ´ar, and R. B. Girshick, “Segment anything,”2023 IEEE/CVF International Conference on Computer Vision (ICCV), pp. 3992–4003, 2023. [Online]. Available: https://api.semanticscholar.org/CorpusID:257952310

    2023

  3. [2]

    Emerging properties in self-supervised vision transformers,

    M. Caron, H. Touvron, I. Misra, H. J’egou, J. Mairal, P. Bojanowski, and A. Joulin, “Emerging properties in self-supervised vision transformers,”2021 IEEE/CVF International Conference on Computer Vision (ICCV), pp. 9630–9640, 2021. [Online]. Available: https: //api.semanticscholar.org/CorpusID:233444273

    2021

  4. [3]

    Sim ´eoni, H

    O. Sim ´eoni, H. V . V o, M. Seitzer, F. Baldassarre, M. Oquab, C. Jose, V . Khalidov, M. Szafraniec, S. Yi, M. Ramamonjisoa, F. Massa, D. Haziza, L. Wehrstedt, J. Wang, T. Darcet, T. Moutakanni, L. Sentana, C. Roberts, A. Vedaldi, J. Tolan, J. Brandt, C. Couprie, J. Mairal, H. J ´egou, P. Labatut, and P. Bojanowski, “DINOv3,” 2025. [Online]. Available: h...

    arXiv 2025

  5. [4]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark, G. Krueger, and I. Sutskever, “Learning transferable visual models from natural language supervision,” inInternational Conference on Machine Learning,

  6. [5]

    Lora: Low-rank adaptation of large language models,

    E. J. Hu, Y . Shen, P. Wallis, Z. Allen-Zhu, Y . Li, S. Wang, L. Wang, and W. Chen, “Lora: Low-rank adaptation of large language models,”

  7. [6]

    Parameter-efficient transfer learning for nlp,

    N. Houlsby, A. Giurgiu, S. Jastrzebski, B. Morrone, Q. de Laroussilhe, A. Gesmundo, M. Attariyan, and S. Gelly, “Parameter-efficient transfer learning for nlp,” 2019. [Online]. Available: https://arxiv.org/abs/1902. 00751

    2019

  8. [7]

    Available: https://arxiv.org/abs/2106.09685

    [Online]. Available: https://arxiv.org/abs/2106.09685

    arXiv

Show all 93 references
  1. [8]

    ibot: Image bert pre-training with online tokenizer,

    J. Zhou, C. Wei, H. Wang, W. Shen, C. Xie, A. L. Yuille, and T. Kong, “ibot: Image bert pre-training with online tokenizer,”ArXiv, vol. abs/2111.07832, 2021. [Online]. Available: https://api.semanticscholar. org/CorpusID:244117494

    2021 arXiv

  2. [9]

    Sam 2: Segment anything in images and videos,

    N. Ravi, V . Gabeur, Y .-T. Hu, R. Hu, C. Ryali, T. Ma, H. Khedr, R. R¨adle, C. Rolland, L. Gustafsonet al., “Sam 2: Segment anything in images and videos,”arXiv preprint arXiv:2408.00714, 2024

    2024 arXiv

  3. [10]

    Adapting segment anything model for change detection in vhr remote sensing images,

    L. Ding, K. Zhu, D. Peng, H. Tang, K. Yang, and L. Bruzzone, “Adapting segment anything model for change detection in vhr remote sensing images,”IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1–11, 2023. [Online]. Available: https://api.semanticscholar.org/Co...

    2023

  4. [11]

    Integrating sam with feature interaction for remote sensing change detection,

    D. Zhang, F. Wang, L. Ning, Z. Zhao, J. Gao, and X. Li, “Integrating sam with feature interaction for remote sensing change detection,”IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1–11, 2024

    2024

  5. [12]

    Adapting vision transformer for efficient change detection,

    Y . Zhao, Y . Zhang, Y . Dong, and B. Du, “Adapting vision transformer for efficient change detection,”ArXiv, vol. abs/2312.04869, 2023. [Online]. Available: https://api.semanticscholar.org/CorpusID:266149789

    2023 arXiv

  6. [13]

    Integrating deep change vector analysis and sam for class-specific change detection,

    S. Saha and K. Awadhiya, “Integrating deep change vector analysis and sam for class-specific change detection,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, pp. 1–12, 2025

    2025

  7. [14]

    A spatial-temporal attention-based method and a new dataset for remote sensing image change detection,

    H. Chen and Z. Shi, “A spatial-temporal attention-based method and a new dataset for remote sensing image change detection,” vol. 12, no. 10, p. 1662. [Online]. Available: https://www.mdpi.com/2072-4292/ 12/10/1662

  8. [15]

    Change dino: A unified transformer-based framework for object-level change detection and segmentation in remote sensing imagery,

    Y . Chen, R. Zhang, X. Ning, H. Zhang, Y . He, Y . Xie, and J. Wang, “Change dino: A unified transformer-based framework for object-level change detection and segmentation in remote sensing imagery,”IGARSS 2024 - 2024 IEEE International Geoscience and Remote Sensing Symposium,...

    2024

  9. [16]

    Remote sensing image change detection with transformers,

    H. Chen, Z. Qi, and Z. Shi, “Remote sensing image change detection with transformers,” vol. 60, pp. 1–14. [Online]. Available: http://arxiv.org/abs/2103.00208

    arXiv

  10. [17]

    A deeply supervised attention metric-based network and an open aerial image dataset for remote sensing change detection,

    Q. Shi, M. Liu, S. Li, X. Liu, F. Wang, and L. Zhang, “A deeply supervised attention metric-based network and an open aerial image dataset for remote sensing change detection,” vol. 60, pp. 1–16. [Online]. Available: https://ieeexplore.ieee.org/document/9467555/

  11. [18]

    Hybrid attention-aware transformer network collaborative multiscale feature alignment for building change detection,

    C. Xu, Z. Ye, L. Mei, H. Yu, J. Liu, Y . Yalikun, S. Jin, S. Liu, W. Yang, and C. Lei, “Hybrid attention-aware transformer network collaborative multiscale feature alignment for building change detection,”IEEE Transactions on Instrumentation and Measurement, vol. 73, pp. 1– 14...

    2024

  12. [19]

    Transition is a process: Pair-to-video change detection networks for very high resolution remote sensing images,

    M. Lin, G. Yang, and H. Zhang, “Transition is a process: Pair-to-video change detection networks for very high resolution remote sensing images,” vol. 32, pp. 57–71. [Online]. Available: https://ieeexplore.ieee.org/document/9975266/

  13. [20]

    Joint variation learning of fusion and difference features for change detection in remote sensing images,

    K. Jiang, W. Zhang, J. Liu, F. Liu, and L. Xiao, “Joint variation learning of fusion and difference features for change detection in remote sensing images,”IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1–18, 2022. [Online]. Available: https://api.semanticsch...

    2022

  14. [21]

    Z. Li, C. Yan, Y . Sun, and Q. Xin, “A densely attentive refinement network for change detection based on very-high-resolution bitemporal JOURNAL OF LATEX CLASS FILES, VOL. 18, NO. 9, SEPTEMBER 2025 15 remote sensing images,” vol. 60, pp. 1–18. [Online]. Available: https://iee...

    2025

  15. [22]

    Bitemporal attention sharing network for remote sensing image change detection,

    Z. Wang, G. Gu, M. Xia, L. Weng, and K. Hu, “Bitemporal attention sharing network for remote sensing image change detection,” vol. 17, pp. 10 368–10 379

  16. [23]

    Robust change detection for remote sensing images based on temporospatial interactive attention module,

    J. Wei, K. Sun, W. Li, W. Li, S. Gao, S. Miao, Q. Zhou, and J. Liu, “Robust change detection for remote sensing images based on temporospatial interactive attention module,” vol. 128, p. 103767

  17. [24]

    Agformer: An anchor- guided transformer for class imbalance in remote sensing change detection,

    J. Chen, D. Wu, Q. Ma, S. Xu, and Y . Zheng, “Agformer: An anchor- guided transformer for class imbalance in remote sensing change detection,”Pattern Recognit., vol. 168, p. 111839, 2025. [Online]. Available: https://api.semanticscholar.org/CorpusID:278991383

    2025

  18. [25]

    Changeclip: Remote sensing change detection with multimodal vision-language representation learn- ing,

    S. Dong, L. Wang, B. Du, and X. Meng, “Changeclip: Remote sensing change detection with multimodal vision-language representation learn- ing,”ISPRS Journal of Photogrammetry and Remote Sensing, vol. 208, pp. 53–69, 2024

    2024

  19. [26]

    Changemamba: Re- mote sensing change detection with spatiotemporal state space model,

    H. Chen, J. Song, C. Han, J. Xia, and N. Yokoya, “Changemamba: Re- mote sensing change detection with spatiotemporal state space model,” IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1– 20, 2024

    2024

  20. [27]

    Exchanging dual- encoder–decoder: A new strategy for change detection with semantic guidance and spatial localization,

    S. Zhao, X.-l. Zhang, P. Xiao, and G. He, “Exchanging dual- encoder–decoder: A new strategy for change detection with semantic guidance and spatial localization,” vol. 61, pp. 1–16

  21. [28]

    Difference-aware multiscale feature aggregation network for building change detection,

    T. Zhan, Q. Tian, Y . Zhu, J. Lan, Q. Dang, and M. Gong, “Difference-aware multiscale feature aggregation network for building change detection,”IEEE Transactions on Geoscience and Remote Sensing, vol. 63, pp. 1–15, 2025. [Online]. Available: https: //api.semanticscholar.org/C...

    2025

  22. [29]

    Difference enhancement and dependency- aware network for remote sensing images change detection,

    H. Li, S. Cheng, and A. Du, “Difference enhancement and dependency- aware network for remote sensing images change detection,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2025. [Online]. Available: https://api.semanticscholar. org/CorpusID:...

    2025

  23. [30]

    Dual- granularity feature alignment for change detection in remote sensing images,

    F. Zhou, X. Zhang, H. Shuai, R. Hang, S. Zhu, and T. Geng, “Dual- granularity feature alignment for change detection in remote sensing images,” vol. 18, pp. 4487–4497

  24. [31]

    Overcoming the uncertainty challenges in detecting building changes from remote sensing images,

    J. Li, W. He, Z. Li, Y . Guo, and H. Zhang, “Overcoming the uncertainty challenges in detecting building changes from remote sensing images,” vol. 220, pp. 1–17. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S092427162400426X

  25. [32]

    Cwmamba: Leveraging cnn-mamba fusion for enhanced change detection in remote sensing images,

    Y . Liu, G. Cheng, Q. Sun, C. Tian, and L. Wang, “Cwmamba: Leveraging cnn-mamba fusion for enhanced change detection in remote sensing images,”IEEE Geoscience and Remote Sensing Letters, vol. 22, pp. 1–5, 2025. [Online]. Available: https://api.semanticscholar. org/CorpusID:276819008

    2025

  26. [33]

    FTA-net: Frequency-temporal-aware network for remote sensing change detection,

    T. Zhu, Z. Zhao, M. Xia, J. Huang, L. Weng, K. Hu, H. Lin, and W. Zhao, “FTA-net: Frequency-temporal-aware network for remote sensing change detection,” vol. 18, pp. 3448–3460

  27. [34]

    Depthcd: Depth prompting in 2d remote sensing imagery change detection,

    N. Zhou, M. Zhou, and H. Sui, “Depthcd: Depth prompting in 2d remote sensing imagery change detection,”ISPRS Journal of Photogrammetry and Remote Sensing, 2025. [Online]. Available: https://api.semanticscholar.org/CorpusID:279383174

    2025

  28. [35]

    Sam-mamba:a two-stage change detection network combining the adapting segment anything and mamba models,

    Y . Li, W. Liu, E. Li, L. Zhang, and X. Li, “Sam-mamba:a two-stage change detection network combining the adapting segment anything and mamba models,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2025. [Online]. Available: https://api.semant...

    2025

  29. [36]

    A deeply supervised image fusion network for change detection in high resolution bi-temporal remote sensing images,

    C. Zhang, P. Yue, D. Tapete, L. Jiang, B. Shangguan, L. Huang, and G. Liu, “A deeply supervised image fusion network for change detection in high resolution bi-temporal remote sensing images,”Isprs Journal of Photogrammetry and Remote Sensing, vol. 166, pp. 183–200,

  30. [37]

    Cd-stmamba: Toward remote sensing image change detection with spatio-temporal interaction mamba model,

    S. Liu, S. Wang, W. Zhang, T. Zhang, M. Xu, M. Yasir, and S. Wei, “Cd-stmamba: Toward remote sensing image change detection with spatio-temporal interaction mamba model,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 18, pp. 10 471–10 48...

    2025

  31. [38]

    Hcgmnet: A hierarchical change guiding map network for change detection,

    C. Han, C. Wu, and B. Du, “Hcgmnet: A hierarchical change guiding map network for change detection,”IGARSS 2023 - 2023 IEEE International Geoscience and Remote Sensing Symposium, pp. 5511–5514, 2023. [Online]. Available: https://api.semanticscholar.org/ CorpusID:257050289

    2023

  32. [39]

    Change guiding network: Incorporating change prior to guide change detection in remote sensing imagery,

    C. Han, C. Wu, H. Guo, M. Hu, J. Li, and H. Chen, “Change guiding network: Incorporating change prior to guide change detection in remote sensing imagery,” vol.{}, pp. 1–17

  33. [41]

    Beyond cross-temporal difference: Style-aligned and fusion-difference learning for change detection,

    S. Fu, S. Dong, and X. Meng, “Beyond cross-temporal difference: Style-aligned and fusion-difference learning for change detection,” IEEE Transactions on Geoscience and Remote Sensing, 2025. [Online]. Available: https://api.semanticscholar.org/CorpusID:279703915

    2025

  34. [42]

    Feature hierarchical differentiation for remote sensing image change detection,

    G. Pei and L. Zhang, “Feature hierarchical differentiation for remote sensing image change detection,” vol. 19, pp. 1–5. [Online]. Available: https://ieeexplore.ieee.org/document/9837915/

  35. [43]

    Efficientcd: A new strategy for change detection based with bi-temporal layers exchanged,

    S. Dong, Y . Zhu, G. Chen, and X. Meng, “Efficientcd: A new strategy for change detection based with bi-temporal layers exchanged,”IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1– 13, 2024. [Online]. Available: https://api.semanticscholar.org/CorpusID: 271334086

    2024

  36. [44]

    Change detection on remote sensing images using dual-branch multilevel intertemporal network,

    Y . Feng, J. Jiang, H. Xu, and J. Zheng, “Change detection on remote sensing images using dual-branch multilevel intertemporal network,” vol. 61, pp. 1–15

  37. [45]

    Aegl-net: Adaptive multiscale global–local feature fusion network for remote sensing change detection,

    Z. Ying, Y . Zhou, Y . Zhai, H. Zhu, H. Zhang, P. Coscia, A. Genovese, F. Scotti, V . Piuri, and C. L. P. Chen, “Aegl-net: Adaptive multiscale global–local feature fusion network for remote sensing change detection,”IEEE Transactions on Geoscience and Remote Sensing, vol. 63, ...

    2025

  38. [46]

    Adapting segment anything model for change detection in vhr remote sensing images,

    L. Ding, K. Zhu, D. Peng, H. Tang, K. Yang, and L. Bruzzone, “Adapting segment anything model for change detection in vhr remote sensing images,”IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1–11, 2024

    2024

  39. [47]

    SGANet: A siamese geometry-aware network for remote sensing change detection,

    J. Chen, S. Dong, and X. Meng, “SGANet: A siamese geometry-aware network for remote sensing change detection,” vol. 18, pp. 6232–6248

  40. [48]

    Changer: Feature interaction is what you need for change detection,

    S. Fang, K. Li, and Z. Li, “Changer: Feature interaction is what you need for change detection,” vol. 61, pp. 1–11

  41. [49]

    Hfifnet: Hierarchical feature interaction network with multiscale fusion for change detection,

    M. Han, T. Xu, Q. Liu, X. Yang, J. Wang, and J. Kong, “Hfifnet: Hierarchical feature interaction network with multiscale fusion for change detection,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 18, pp. 4318–4330,

  42. [50]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,”2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 770–778, 2015. [Online]. Available: https://api.semanticscholar.org/CorpusID:206594692

    2016

  43. [51]

    A transformer-based siamese net- work for change detection,

    W. G. C. Bandara and V . M. Patel, “A transformer-based siamese net- work for change detection,” inIGARSS 2022-2022 IEEE International Geoscience and Remote Sensing Symposium. IEEE, 2022, pp. 207–210

    2022

  44. [52]

    Fccdn: Feature constraint network for vhr image change detection,

    P. Chen, D. Hong, Z. Chen, X. S. Yang, B. Li, and B. Zhang, “Fccdn: Feature constraint network for vhr image change detection,”ArXiv, vol. abs/2105.10860, 2021. [Online]. Available: https://api.semanticscholar. org/CorpusID:235166250

    2021 arXiv

  45. [53]

    Convformer-cd: Hybrid cnn–transformer with temporal attention for detecting changes in remote sensing imagery,

    F. Yang, M. Li, W. Shu, A. Qin, T. Song, C. Gao, and G.-S. Xia, “Convformer-cd: Hybrid cnn–transformer with temporal attention for detecting changes in remote sensing imagery,”IEEE Transactions on Geoscience and Remote Sensing, vol. 63, pp. 1–15, 2025. [Online]. Available: htt...

    2025

  46. [54]

    Eatder: Edge-assisted adaptive transformer detector for remote sensing change detection,

    J. Ma, J. Duan, X. Tang, X. Zhang, and L. Jiao, “Eatder: Edge-assisted adaptive transformer detector for remote sensing change detection,” IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1– 15, 2024. [Online]. Available: https://api.semanticscholar.org/CorpusI...

    2024

  47. [55]

    Network and dataset for multiscale remote sensing image change detection,

    S. Liu, D. Zhao, Y . Zhou, Y . Tan, H. He, Z. Zhang, and L. Tang, “Network and dataset for multiscale remote sensing image change detection,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 18, pp. 2851–2866,

  48. [56]

    Full-scale change detection network for remote sensing images based on deep feature fusion,

    ——, “Full-scale change detection network for remote sensing images based on deep feature fusion,”IEEE Transactions on Geoscience JOURNAL OF LATEX CLASS FILES, VOL. 18, NO. 9, SEPTEMBER 2025 16 and Remote Sensing, vol. 63, pp. 1–13, 2025. [Online]. Available: https://api.semant...

    2025

  49. [57]

    Aanet: An ambiguity-aware network for remote-sensing image change detection,

    R. Hang, S. Xu, P. Yuan, and Q. Liu, “Aanet: An ambiguity-aware network for remote-sensing image change detection,”IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1–11, 2024. [Online]. Available: https://api.semanticscholar.org/CorpusID:268204920

    2024

  50. [58]

    Sam-based efficient feature integration network for remote sensing change detection: A case study on macao sea reclamation,

    J. Huang, J. Bao, M. Xia, and X. Yuan, “Sam-based efficient feature integration network for remote sensing change detection: A case study on macao sea reclamation,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 18, pp. 16 916–16 928,

  51. [59]

    Feature pyramid networks for object detection,

    T.-Y . Lin, P. Dollar, R. Girshick, K. He, B. Hariharan, and S. Belongie, “Feature pyramid networks for object detection,” in2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, pp. 936–944. [Online]. Available: http://ieeexplore.ieee.org/ document/8099589/

  52. [60]

    Available: https://api.semanticscholar.org/CorpusID: 275032911

    [Online]. Available: https://api.semanticscholar.org/CorpusID: 275032911

  53. [61]

    CDMamba: Incorporating local clues into mamba for remote sensing image binary change detection,

    H. Zhang, K. Chen, C. Liu, H. Chen, Z. Zou, and Z. Shi, “CDMamba: Incorporating local clues into mamba for remote sensing image binary change detection,” vol. 63, pp. 1–16

  54. [62]

    Interactive and supervised dual-mode attention network for remote sensing image change detection,

    H. Ren, M. Xia, L. Weng, H. Lin, J. Huang, and K. Hu, “Interactive and supervised dual-mode attention network for remote sensing image change detection,” vol. 63, pp. 1–18

  55. [63]

    ELGC- net: Efficient local–global context aggregation for remote sensing change detection,

    M. Noman, M. Fiaz, H. Cholakkal, S. Khan, and F. S. Khan, “ELGC- net: Efficient local–global context aggregation for remote sensing change detection,” vol. 62, pp. 1–11

  56. [64]

    Available: https://api.semanticscholar.org/CorpusID: 279733460

    [Online]. Available: https://api.semanticscholar.org/CorpusID: 279733460

  57. [65]

    A deep supervised change detection network based on context-rich information,

    Y . Zhang, L. Yang, L. Zhu, C. Zhou, Y . A. Nanehkaran, and J. Wang, “A deep supervised change detection network based on context-rich information,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2025. [Online]. Available: https://api.semantic...

    2025

  58. [66]

    Cross attention is all you need: relational remote sensing change detection with transformer,

    K. Lu, X. Huang, R. Xia, P. Zhang, and J. Shen, “Cross attention is all you need: relational remote sensing change detection with transformer,” GIScience & Remote Sensing, vol. 61, 2024. [Online]. Available: https://api.semanticscholar.org/CorpusID:271326551

    2024

  59. [67]

    Remote sensing change detection with transformers trained from scratch,

    M. Noman, M. Fiaz, H. Cholakkal, S. Narayan, R. M. Anwer, S. H. Khan, and F. S. Khan, “Remote sensing change detection with transformers trained from scratch,”IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1–14, 2023. [Online]. Available: https://api.semantic...

    2023

  60. [68]

    Sffce-cd: Spatial and frequency feature cross enhancement for change detection,

    Y . Xing, J. Hu, B. Jiang, Q. Zhao, L. Feng, and R. Huang, “Sffce-cd: Spatial and frequency feature cross enhancement for change detection,” inIEEE International Conference on Acoustics, Speech, and Signal Processing, 2025. [Online]. Available: https: //api.semanticscholar.org...

    2025

  61. [69]

    DSFI-CD: Diffusion- guided spatial-frequency-domain information interaction for remote sensing image change detection,

    X. Li, Y . Tan, K. Liu, X. Wang, and X. Zhou, “DSFI-CD: Diffusion- guided spatial-frequency-domain information interaction for remote sensing image change detection,” vol. 63, pp. 1–18

  62. [70]

    Global- aware siamese network for change detection on remote sensing images,

    R. Zhang, H. Zhang, X. Ning, X. Huang, J. Wang, and W. Cui, “Global- aware siamese network for change detection on remote sensing images,” vol. 199, pp. 61–72

  63. [71]

    Mfatnet: Multi-scale feature aggregation via transformer for remote sensing image change detection,

    Z. Mao, X.-Y . Tong, Z. Luo, and H. Zhang, “Mfatnet: Multi-scale feature aggregation via transformer for remote sensing image change detection,”Remote. Sens., vol. 14, p. 5379, 2022. [Online]. Available: https://api.semanticscholar.org/CorpusID:253220641

    2022

  64. [72]

    MLDFNet: A multilabel dual- flow network for change detection in bitemporal remote sensing images,

    D. Sidekejiang, P. Zheng, and L. Wang, “MLDFNet: A multilabel dual- flow network for change detection in bitemporal remote sensing images,” vol. 18, pp. 4867–4880

  65. [73]

    B2cnet: A progressive change boundary-to-center refinement network for multitemporal remote sensing images change detection,

    Z. Zhang, L. Bao, S. Xiang, G. Xie, and R. Gao, “B2cnet: A progressive change boundary-to-center refinement network for multitemporal remote sensing images change detection,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 17, pp. 11 322–1...

    2024

  66. [74]

    Dual fine-grained network with frequency transformer for change detection on remote sensing images,

    Z. Li, Z. Zhang, M. Li, L. Zhang, X. Peng, R. He, and L. Shi, “Dual fine-grained network with frequency transformer for change detection on remote sensing images,” vol. 136, p. 104393. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S1569843225000408

  67. [75]

    Stransunet: A siamese transunet- based remote sensing image change detection network,

    J. Yuan, L. Wang, and S. Cheng, “Stransunet: A siamese transunet- based remote sensing image change detection network,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 15, pp. 9241–9253, 2022. [Online]. Available: https://api.semanticschol...

    2022

  68. [76]

    HASNet: A foreground association-driven siamese network with hard sample opti- mization for remote sensing image change detection,

    C. Tao, D. Kuang, Z. Huang, C. Peng, and H. Li, “HASNet: A foreground association-driven siamese network with hard sample opti- mization for remote sensing image change detection,” vol. 63, pp. 1–16

  69. [77]

    Fully convolutional networks for multisource building extraction from an open aerial and satellite imagery data set,

    S. Ji, S. Wei, and M. Lu, “Fully convolutional networks for multisource building extraction from an open aerial and satellite imagery data set,”IEEE Transactions on Geoscience and Remote Sensing, vol. 57, pp. 574–586, 2019. [Online]. Available: https: //api.semanticscholar.org...

    2019

  70. [78]

    An attention-based multiscale transformer network for remote sensing image change detection,

    W. Liu, Y .-Y . Lin, W. Liu, Y . Yu, and J. Li, “An attention-based multiscale transformer network for remote sensing image change detection,”ISPRS Journal of Photogrammetry and Remote Sensing,

  71. [79]

    S2looking: A satellite side-looking dataset for building change detection,

    L. Shen, Y . Lu, H. Chen, H. Wei, D. Xie, J. Yue, R. Chen, Y . Zhang, A. Zhang, S. Lv, and B. Jiang, “S2looking: A satellite side-looking dataset for building change detection,”Remote. Sens., vol. 13, p. 5094,

  72. [80]

    Rs-mamba for large remote sensing image dense prediction,

    S. Zhao, H. Chen, X. Zhang, P. Xiao, L. Bai, and W. Ouyang, “Rs-mamba for large remote sensing image dense prediction,”IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1–14, 2024

    2024

  73. [81]

    Remote sensing image change detection transformer network based on dual-feature mixed attention,

    X. Song, Z. Hua, and J. Li, “Remote sensing image change detection transformer network based on dual-feature mixed attention,”IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1– 16, 2022. [Online]. Available: https://api.semanticscholar.org/CorpusID: 252605199

    2022

  74. [82]

    Transunetcd: A hybrid transformer network for change detection in optical remote-sensing images,

    Q. Li, R. Zhong, X. Du, and Y . Du, “Transunetcd: A hybrid transformer network for change detection in optical remote-sensing images,”IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1–19, 2022

    2022

  75. [83]

    Lccdmamba: Visual state space model for land cover change detection of vhr remote sensing images,

    J. Huang, X. Yuan, C. T. Lam, Y . Wang, and M. Xia, “Lccdmamba: Visual state space model for land cover change detection of vhr remote sensing images,”IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 18, pp. 5765–5781,

  76. [84]

    Available: https://api.semanticscholar.org/CorpusID: 275776209

    [Online]. Available: https://api.semanticscholar.org/CorpusID: 275776209

  77. [86]

    Change detection in remote sensing images using conditional adversarial networks,

    M. Lebedev, Y . V . Vizilter, O. Vygolov, V . A. Knyaz, and A. Y . Rubis, “Change detection in remote sensing images using conditional adversarial networks,”The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences,

  78. [89]

    Available: https://api.semanticscholar.org/CorpusID: 236134438

    [Online]. Available: https://api.semanticscholar.org/CorpusID: 236134438

  79. [91]

    Swinsunet: Pure transformer network for remote sensing image change detection,

    C. Zhang, L. Wang, S. Cheng, and Y . Li, “Swinsunet: Pure transformer network for remote sensing image change detection,”IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1–13, 2022

    2022

  80. [93]

    Dc-mamba: A novel network for enhanced remote sensing change detection in difficult cases,

    J. Zhang, R. Chen, F. Liu, H. Liu, B. Zheng, and C. Hu, “Dc-mamba: A novel network for enhanced remote sensing change detection in difficult cases,”Remote Sensing, vol. 16, no. 22, 2024. [Online]. Available: https://www.mdpi.com/2072-4292/16/22/4186

    2024

  81. [2018]

    Available: https://api.semanticscholar.org/CorpusID: 57660599

    [Online]. Available: https://api.semanticscholar.org/CorpusID: 57660599

  82. [2020]

    Available: https://api.semanticscholar.org/CorpusID: 225504855

    [Online]. Available: https://api.semanticscholar.org/CorpusID: 225504855

  83. [2021]

    Available: https://api.semanticscholar.org/CorpusID: 231591445

    [Online]. Available: https://api.semanticscholar.org/CorpusID: 231591445

  84. [2023]

    Available: https://api.semanticscholar.org/CorpusID: 259931492

    [Online]. Available: https://api.semanticscholar.org/CorpusID: 259931492

  85. [2025]

    Available: https://api.semanticscholar.org/CorpusID: 275486271

    [Online]. Available: https://api.semanticscholar.org/CorpusID: 275486271

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.